morph algorithms on gpus - texas state university.pdfVIP

Morph Algorithms on GPUs Rupesh Nasre Martin Burtscher Keshav Pingali Inst. for Computational Engineering and Sciences, University of Texas at Austin, USA Dept. of Computer Science, Texas State University, San Marcos, USA Dept. of Computer Science, University of Texas at Austin, USA nasre@ices.utexas.edu, burtscher@txstate.edu, pingali@cs.utexas.edu Abstract 1. Introduction There is growing interest in using GPUs to accelerate graph al- Graphics Processing Units (GPUs) have been shown to be very use- gorithms such as breadth-first search, computing ranks, and ful in many application domains outside of graphics. GPU hard- finding shortest paths. However, these algorithms do not modify ware is designed to process blocks of pixels at high speed and with the graph structure, so their implementation is relatively easy com- wide parallelism, so it is well suited for executing regular algo- pared to general graph algorithms like mesh generation and refine- rithms that operate on dense vectors and matrices. We understand ment, which morph the underlying graph in non-trivial ways by much less about how to use GPUs effectively to execute irregu- adding and removing nodes and edges. We know relatively little lar algorithms that use dynamic data structures like graphs and about how to implement morph algorithms efficiently on GPUs. trees. Harish and Narayanan [10] pioneered this field with their In this paper, we present and study four morph algorithms: